ABSTRACT
There is a lack of studies on whether the use of open source mid-resolution image and traditional classification techniques can produce a high-quality output with sufficient level of accuracy required to map mangrove forest trees at species level. To bridge this gap, we employed maximum likelihood algorithm to identify and map out mangrove species composition using open source mid-resolution Landsat data, taking Bangladesh Sundarbans as a case study. Our results showed that such mid-resolution image can yield acceptable accuracy levels to be used in the image analysis at species level. We achieved an overall accuracy of 89.10% and kappa coefficient of 0.87 for the five-identified species, viz. Heritiera fomes, Ceriops decandra, Excoecaria agallocha, Sonneratia apetala, and Xylocarpus mekongensis, which is higher than the required minimum overall accuracy of 85% deemed suitable to use in most of the natural resource mapping applications. With this accuracy level, the spatial coverage of the investigated species for a single year was calculated. We concluded that mid-resolution images, such as Landsat, and the traditional classification algorithm can be applied with confidence for the identification and classification of mangrove forest resources at species level as an alternative to the high resolution satellite images.
